Académique Documents
Professionnel Documents
Culture Documents
OSKAR LARSSON
Reliability as a concept
General definition
The ability to meet specific requirements under a specified
period
2
Background
• During the second world war it became evident that the
reliability of complex technical installations was a problem
• As an example the modern warships at the time were
only operational for attack/defence in about 60 % of the
time
• Similar effects were observed on the reliability of e.g. the
V1l V2 rocket systems -the first of many launches were
unsuccessful
• These events were the real initiators of the reliability
theory for technical components and systems
3
Background
• lnitially the reliability theory was developed for systems
with a large number of {semi-} identical components
subject to the same exposure conditions
– electrical systems {bulbs, switches, .. }
• and later on to
– nuclear power installations {valves, pipes, pumps, .. }
– chemical plants (pipes, pressure vessels, valves,
pumps, .. }
– manufacturing plants (pumps, compressors,
conveyers, .. }
4
Background
• For such systems the probability of a
component failure may be assessed in a
frequentistic manner – from observed failure
rates – number of failures per component
operation hours
• Due to the characteristics of the failure
mechanism there is a steady deterioration as
a function of time/use
• For the considered components the main
concern was centered around the statistical
modeling of time until failure
5
Background
• For technical systems such as structures and structural
components the classical reliability analysis is of limited
use because
– all structural components are unique
– the failure mechanism tends to be related to extreme
load events exceeding the residual capacity of the
component- not the direct effect of deterioration alone
• For such systems a different approach is thus required,
namely
– an individual modeling of both the resistance as a
function of time and the loading as a function of time
6
Connection to previous lectures
• Basic statistics:
– Distributions, (mean, std. deviation, parameters)
– Probability of failure
– Standard normal distribution
• Introduction
– Codes and standards, partial coefficients
– Limit states
– Modeling of uncertainties
7
Methods to measure reliability
• Level I method
Uncertain parameters are modelled by one characteristic
value, e.g. code based on partial coefficent method
• Level II method
Uncertain parameters modelled by the mean values and
the standard deviations and by the correlation coefficietns
between stochastic vairables. The stochastic variables are
implicitly assumed to be normally distributed. The reliability
index method is an example of a level II method
8
Methods to measure reliability
• Level III method
Uncertain quantities are modelled by their joint distribution
functions. The probability of failure is estimated as a
measure of the reliability
• Level IV method
Consequences (cost) of failure are also taken into account
9
Deterministic reliability measures
(Level I)
10
Safety factor
• Safety factor is a traditional measure on reliability
11
Partial safety factor format
Rk
Rd (Gk G Qk Q ) S d
R
Index k denotes characteristic value
Characteristic values are defined in statistical terms
i denote partial safety factors reflecting different degree of
uncertainty for basic parameters
12
Implicit and generic risk analysis
behind modern structural codes
• Consequences and risk acceptance in the society are
implicitly considered in modern structural codes
• The results are presented in terms of required levels of
formal reliability in different situations
• This means that the decision problem when designing a
structure is transformed to a verification that the
probability of violation of the limit state is lower that a
prescribed level.
• Modern structural codes are often called limit state
design codes
13
Probabilistic reliability measures
14
Structural reliability
Reliability of structures cannot be assessed
through failure rates because:
15
Probability of failure, reliability
Probability of failure pf = P(R-S<0)
Reliability = 1- pf
16
Main steps in reliability analysis
1. Select a target reliability level (safety or consequence
class)
2. Identify significant failure modes (deflection, bending)
3. Formulate limit state functions (g(E,R) = MEd – MRd = 0)
4. Identify stochastic variables and deterministic
parameters. Specify distribution types and statistical
parameters
5. Estimate reliability of each failure mode
6. Change design if reliabiliy does not meat target reliabilty
7. Evaluate reliability result by a sensitivity analysis
17
Structural Engineering Decision
Problem:
Optimal Design:
Expected Benefit of the Structure
Benefit of the Structure in Service
Reliability Risk
E B
E B I 1 PF CD CD CF PF CD 0
CD
Benefit
But what is a proper
target reliability???
Risk
18
Cost
Theory of Structural
Reliability
19
Failure
In general defined as: capacity < demand
resistance < load effect
Gl
f mW r s
4
BUT - resistance and load effect are uncertain !!!
First assume that r and s are Normal distributed
random variables with known mean values and
standard deviations.
20
Probability of failure
The probability, that
resistance < load effect
Gl
R f mW ; S
4
PF P R S P R S 0
S ~ N S , S R ~ N R , R
21
Probability of failure
The probability, that
resistance < load effect
Gl
R f mW ; S Definition:
4 Safety margin
PF P R S P R S 0 M=R-S
Probability Density
S ~ N S , S R ~ N R , R
M ~ N M , M
M R S
M2 R2 S2
0
0 M
PF f M m dm
M
0
0 M
PF f M m dm
M
Reliability Index
The probability, that
24
Reliability Index (Safety index)
M
PF
M
25
PF
Reliability Index 0
1
0.5
0.158655
2 0.02275
M
PF
3 0.00135
M 4 3.17E‐05
5 2.87E‐07
6 9.87E‐10
Geometrical interpretation:
7 1.28E‐12
8 6.22E‐16
Probability Density
26
Safety class
27
Summary
If R and S are Normal distributed
the safety margin M is also Normal M=R–S
Distributed
The probability of failure is then PF = P(R-S<0) = P(M<0)
With mean value of M μM = μR - μS
0
The probability of failure Ф Ф
Q Q
Q μ = 30 kN COV = 20 %
29
Reliability Analysis, the general case
Limit state function general case
In the general case the resistance and the load may be defined in
terms of functions where X are basic random variables
R = f1(X), S = f2(X)
M= R-S = f1(X)-f2(X) = g(X)
30
Illustration for the general case with
random variables
31
Probability of failure Pf
Pf =P(M0) = P(g(X)0) X = (x1,.....xn )
Then
Pf f X ( X )dX
f
where f is the failure region defined by
the limit state function
32
Reliability Analysis, the general case
33
Analysis methods in reliability analysis
• Analytical methods (only in very special cases)
• First Order Reliability Methods (FORM)
• Second Order Reliability Methods (SORM)
• Simulation techniques
34
Special case: linear limit state function with normal
distributed, independent random variables
M g ( X 1 ... X n ) a 0 a 1 X 1 ... a n X n
M a0 a1 x1 ....an xn
36
Reliability analysis , the general case
X X
U Standard Normal
X Y 0
Y 1 Scaling
Normal
Shift
X
37
Non-linear failure function
The failure function can be linearized by Taylor expansion:
n
g
*
M g (u ) U i
(U i u * )
1 u u *
39
Geometrical interpretation
Usually the limit state function is
non-linear
- this small phenomenon caused
the so-called invariance problem
40
Iteration process
The optimization problem can be
formulatedas an iteration problem
42
Example 2
Consider the steel rod with cross- h = r ·a
sectional area a and yield stress r
43
Example 2
The initial step is to transform the basic random variables into
standardized Normal distributed space.
44
Example 2
Choose an arbitrary first design point u = βα
g(u) = 0
This gives
2000
350 350 300 35
45
Example 2
Normal vector at this point
∙ 350 35 ∙ 350 35
∙ 300
Then
350 35
350 35 350 35 300
350 35
350 35 350 35 300
300
350 35 350 35 300 46
Example 2 – Summary
Safety index:
∙ 350 35 ∙ 300
Normal vector
47
Example 2 – Iteration
Start 1 2 3 4 5 6 7
β 3 -3.9604 3.162993 3.764099 3.74638 3.744948 3.744835 3.744826
αR 1 -0.64088 -0.63663 -0.56327 -0.56049 -0.56102 -0.56099 -0.561
αA 1 -0.64088 -0.63663 -0.56327 -0.56049 -0.56102 -0.56099 -0.561
αS 1 0.422556 0.435216 0.604532 0.609681 0.60869 0.608749 0.60874
∂g/duR 455 438.8344 279.5224 275.7932 276.5072 276.4648 276.4712 276.4707
∂g/duA 455 438.8344 279.5224 275.7932 276.5072 276.4648 276.4712 276.4707
∂g/duS -300 -300 -300 -300 -300 -300 -300 -300
48
FORM with non-normal random
variables
An arbitrary distribution function FX with density function f X may
be transformed to an equivalent standardized normal distribution
in a point x so that the distribution function value and the density
function value are the same:
xi i 1 xi i
FX i ( xi ) ( ) f X i ( xi )
i i i
This transformation is made in each iteration step
49
FORM with correlated, non-normal
variables
50
Second order reliability method
(SORM)
• Is a more exact method since the expansion of the non-linear
limit state function also includes the second order derivatives of
g(X)
FORM
51
Comparison FORM - SORM
• The difference is small for cases with large safety
index (or low probability of failure)
• For higher probability of failure the error in the FORM
approximation can be significant
52
Monte Carlo Simulation
• Instead of calculating the reliability index, i.e. the failure probability
analytically – it might be much easier to run virtual experiments
based on the probabilistic models and the limit state function.
Random number
FX i ( xi ) 20
1 18
16
zj
14
xi
xj 12
Resistance
Random outcome of X 10
nf 6
pf 4
m
2
0
53 -2 0 2 4 6 8 10 12
Load
Simulation methods
54